monitoring: search-blitz - add additional graphs to track p90 and p75 latencies of queries

Created by: ggilmore

This PR adds 3 additional graphs to search-blitz to track p90 latencies of sentinel queries. The latencies of successful and failed queries are broken out into separate graphs, along with another graph that aggregates successful+failed queries together.

Questions:

• Should we consider adding more buckets to the histogram? The graphs look pretty choppy - many more granularity would help?

https://github.com/sourcegraph/sourcegraph/blob/82251e9550fd753af44fb2f282c3fe966403549d/cmd/frontend/graphqlbackend/search_results.go#L375-L380

• Why do we use both "latency" and "duration" in graph titles? Compare the following graphs - they're tracking the same metric, they only differ in whether or not they filter out failed requests:

https://github.com/sourcegraph/sourcegraph/blob/82251e9550fd753af44fb2f282c3fe966403549d/monitoring/definitions/frontend.go#L659-L689